Chapters
Claude Code 101
This is my personal course on how I actually build with Claude Code: the whole arc, from installing the agent and making a first commit to running orchestrated builds where several subagents work in parallel behind acceptance gates. Everything in it comes from real sessions on real projects. Start at chapter one, or pick the entry point below that matches where you already are.
Where are you starting from?
- Start hereNever used an AI coding agentStart at the very beginning: install, first session, first commit.→ Your first session
- Start hereYou use Claude Code, but without much processTighten up: context discipline so multi-day work stops rotting.→ Context discipline: multi-day work without rot
- Start hereYou plan well but build serially, no orchestrationRun a real build: TDD, diagnosis loops, parallel subagents behind gates.→ Execution: running a multi-agent build
- Start hereWeatherproofing the workflowPro practices for preventing expensive failures: hooks, headless CI, PR review, scoped subagents.→ Pro practices: hardening the workflow
The chapters
- 1Your first sessionInstall Claude Code, run a first session, make a small real change, and commit it. Along the way, the mental model that makes the tool legible.
- 2Project setup: a repo agents work well inTurn a repo into a place agents work well in. Write a contract that's actually true, encode conventions as mechanical checks, and prune permissions instead of hoarding them.
- 3Context discipline: multi-day work without rotKeep long, multi-day work coherent by moving memory out of the context window and into docs a fresh session can boot from.
- 4Planning: from idea to agent-executable specTurn a vague idea into a plan an agent can execute without guessing. A product conversation, a relentless grilling, then a PRD cut into vertical slices.
- 5Execution: running a multi-agent buildTurn an approved plan into shipped code with TDD, a diagnosis loop, and parallel subagents merged behind acceptance gates.
- 6Pro practices: hardening the workflowFive hardening practices (hooks, headless runs, PR review, scoped subagents, config hygiene), each drawn from a real scar in my own usage history.